Counter example to Eckart–Young theorem

98 Views Asked by At

The Eckart–Young theorem provides for us the unique $k$ rank approximation matrix such that,

$$ A_{k} = \arg \min ||A- B|| $$

and

$$||A- A_{k}||_{2} = \sigma_{k+1} $$

Where $A_{k} = \sum_{n=1}^{k} \sigma_{k}u_{k}v_{k}^{H}$. I seem to have found a counterexample, consider the case $k=1$ and the matrix

$$A_{*} = (\sigma_{1}-\sigma_{2})u_{1}v^{H}_{1}$$

Does this not also result in

$$||A-A_{*}||_{2} = \sigma_{2}? $$